Revisiting the Gravity Model of Migration

Published date01 May 2023
Date01 May 2023
Subject MatterArticles
Revisiting the Gravity
Model of Migration
Mohammad Azeem Khan1, Zeenat Fatima2
and Sumbul Fatima3
The recent global migration pattern indicates the importance of the move-
ment of people from developing countries to developed countries in search of
better economic opportunities. The G20 report of ‘International Migration and
Displacement Trends’ mentions India at the top of the list of highly educated emi-
grants in G20 countries. The current study addresses the endogeneity problem in
the migration determinants and attempts to highlight the major regional and eco-
nomic determinants of migration flow from India to major OECD countries using
the Gravity model of migration. We apply the Prais–Winsten regression method
to address the cross-sectional correlation, while we apply instrumental variable
regression and Hausman–Taylor regression estimation techniques to deal with
the endogeneity issue. The findings reveal that the population of India, distance,
common official language and per capita income differential are the major deter-
minants of migration from India. In the backdrop of our findings, in terms of per
capita income differential, there is a need for an upward revision in the pay scale
of the white-collar workers in the organised sector.
JEL Codes: C23, F22, J11, J60
International migration, population, unemployment, panel data, endogeneity
Theoretically, the gravity model of migration is usually based on the Random
Utility Maximisation (RUM hereafter) model. The RUM model describes the
Foreign Trade Review
58(2) 329–349, 2023
© 2022 Indian Institute of
Foreign Trade
Reprints and permissions:
DOI: 10.1177/00157325221088707
1Department of Economic Sciences, Indian Institute of Technology Kanpur, Uttar Pradesh, India.
2Indian Institute of Foreign Trade, Centre of WTO Studies, New Delhi, India.
3Department Agricultural Economic and Business Management, Aligarh Muslim University, Aligarh,
Uttar Pradesh, India.
Corresponding author:
Mohammad Azeem Khan, Department of Economic Sciences, Indian Institute of Technology Kanpur,
Uttar Pradesh 208016, India.
330 Foreign Trade Review 58(2)
utility an individual gets by living in a particular country compared to the expected
utility of moving to an alternative place. The RUM model assumes that the pull
factors (like better employment opportunities) in the migrants’ destination should
not be affected by the migration flow itself. Ramos (2016), however, believes that
the attractiveness of a region due to its low levels of unemployment compared to
migrants’ origin region stimulates the inflows of immigrants. Consequently, unem-
ployment in this region may increase while it may parallelly decline in the origin
region. Gravity models do not incorporate these second-round repercussions;
hence, they may give misleading conclusions. Such explanatory factors that are not
affected by the current period migration flows but may be affected by the immedi-
ate past periods’ migration flows are called predetermined variables in time series
or panel data settings. These panel data gravity models can be empirically applied
even in the presence of predetermined variables, provided that the migration flow
(i.e., the dependent variable) is not serially correlated. However, several studies
related to migration suggest that past periods’ migration movements affect the cur-
rent flows (e.g., Disney et al., 2015; Epstein & Gang, 2006; Greenwood, 1970;
Zimmer, 2008, among others). Such migration movements might cause the prob-
lem of endogeneity in the above-discussed predetermined variables.
Nonetheless, in an empirical framework, gravity model can be successfully
used after taking due care of the possible endogeneity problem owing to these
second-round effects. Most of the previous works concerning the migration flow,
especially from developing countries to developed countries, do not address the
issue of endogeneity. For example, the population of both the destination and
source countries and the existing diaspora of the migrants’ source country may be
endogenous. The endogeneity occurs either due to the bidirectional causality (for
instance, the populations in both the regions may also be affected by the volume
of migration in that period) or because of omission of some important variables
from the model, which might affect other explanatory variables included in the
model. For instance, the educational status of the migrants, if not included in the
model, might affect the prevailing employment rates difference between the coun-
tries. Such a problem induces biases in the estimates of the model and makes the
drawn conclusion less reliable.
Another critical issue in the panel data models is the cross-sectional (or contem-
poraneous) correlation among the errors of different groups due to the presence of
common shocks or spatial dependence or unobserved factors that are ultimately
subsumed in the errors. If the cross-sectional correlation is due to the common
factors, then applying the traditional fixed effect and random effect models gives
consistent estimates, though the standard errors are biased. The empirical estima-
tion in the present study addresses the problems mentioned above using the Prais–
Winsten regression (Prais & Winsten, 1954) with panel-corrected standard error
(PCSE), instrumental variable regression (IVREG) and Hausman–Taylor regres-
sion (HTREG; Hausman & Taylor, 1981). Contrary to some previous research
works based on gravity models of migration, our findings through these techniques
reveal new explanations to some of the migration determinants and hence fetch
novelty to this work. We find a plethora of empirical studies examining the deter-
minants of migration from various aspects. However, the present findings give a
better picture of unilateral migration flows, which are more prevalent these days.

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT